Intermediaries in the Medicare Market

2024-10-16

2026-03-31

The key outcome of interest is the quality of the insurance plan recommendation from an insurance producer (or insurance agent). Our primary outcome evaluates the quality of an agent's recommendation against what a fully-informed expert would recommend. We are primarily interested in the heterogeneity in recommendations and response rates by insurance agent characteristics based on publicly available data and randomly assigned consumer characteristics such as gender and age. Finally, we test whether randomly assigning consumers to signal that they are soliciting and comparing multiple recommendations improves the quality of agent recommendations.

Insurance agent characteristics we focus on include number of appointments, firm affiliation, number of states licensed in, tenure, and advertising presence. We will also use text analysis from recordings to study variations in the language used by caller/consumer and insurance producer characteristics. We measure recommendation quality as whether or not the agent recommends Medicare Advantage, the average plan price of the recommendation, and the agent’s accuracy when answering factual questions.

We will also look at the number of incorrect/misleading statements an agent makes as a measure of recommendation quality. Our primary outcome looks at the quality of the agent's recommendation against an expert benchmark, but we also use consumer choice, other agent recommendations, and random choice as benchmarks. We will also look at the impact of state-level differences in the regulatory and market environment.

We will conduct a correspondence study that randomly assigns caller and client characteristics to a randomly selected group of insurance agents. To identify insurance agents, we will assemble a list of the most common insurance agents that appear from Google and other public recommendation lists (including government databases, industry affiliation lists, social media websites, and business review and consumer recommendation websites). We will conduct these searches from multiple IP addresses. We will also use text analysis from recordings to study variations in the language used by caller/consumer and insurance producer characteristics.

Not available

Consumers will be randomly assigned to insurance producers by a computer.

Consumers will be randomly assigned to insurance agents (and therefore insurance agent characteristics) by a computer at the individual level. Variation in competition signaling and consumer characteristics such as gender and age will also be randomly assigned.

No

We will collect approximately 400-800 observations, depending on our ability to obtain additional funding and when our current sources of research funding run out. We will exclude incomplete observations, off-topic observations, and responses from individuals who are not licensed insurance producers. (Complete responses from individuals who are not licensed can be reported separately.) We anticipate having funding for a minimum of 400 observations, however there is a chance we will have fewer than 400 observations if response rates are lower than expected or if several observations do not meet our screening criteria.

We will collect approximately 400-800 observations, depending on our ability to obtain additional funding and when our current sources of research funding run out. We will exclude incomplete observations, off-topic observations, and responses from individuals who are not licensed (complete responses from individuals who are not licensed can be reported separately). We anticipate having funding for a minimum of 400 observations, however there is a chance we will have fewer than 400 observations if response rates are lower than expected or if several observations do not meet our screening criteria. The maximum number of observations will also depend on take-up, screening, and whether we are able to obtain additional funding.

We will aim to have a 50/50 balance of the main agent and consumer characteristics of interest (e.g. the gender and age of the consumers) and whether or not consumers signal competition. We do not expect the sample to be perfectly balanced due to anticipated differences in response rates among agents and because the randomization does not ensure exact balance by insurance agent characteristics such as number of states licensed in, number of appointments, etc.